Abstract Details


Name: Subhamoy Chatterjee
Affiliation: Southwest Research Institute
Conference ID: TVS202510162
Title: Exploring the missing link between surface observations and subsurface dynamics through Physics Informed Neural Networks
Authors and Co-Authors: Mausumi Dikpati
Abstract Type: Invited by SOC
Abstract: Observations show that Active Regions (ARs) are not randomly distributed, but instead align along warped toroidal bands (“toroids”) that evolve slowly during the active phase of the solar cycle. Large ARs emerging where the northern and southern toroids are tipped apart in latitude are especially prone to produce major eruptions, particularly if they become magnetically complex. Notably, these tipped configurations often form weeks before the appearance of such ARs, providing an imprint of subsurface states. It is thus imperative to perform an MHD simulation of the subsurface state vectors as proxies for AR emergence. Although the evolution of subsurface state vectors through an MHD model has previously been qualitatively correlated with AR emergence, a data-driven simulation for new AR emergence has not been possible until now. It requires a method to ingest the surface observations and derive the initial configuration state vectors. We here demonstrate a novel Physics Informed Neural Network (PINN) approach that uses the fitted toroids from observed data and a global MHD physical model as constraints to derive the subsurface initial state vectors. We show a hindcast case where the PINN-derived initial state vectors forward modeled through MHD simulation can successfully predict the AR emergence location about a week ahead of actual emergence. Along with current observational capabilities, magnetic proxies from historical long-term datasets such as those from Kodaikanal Observatory can act as a valuable resource to perform an extensive validation of this study.